PRB: Incorrect Results Returned, C Float Functions

• Microsoft FORTRAN for MS-DOS, versions 4.1, 5.0, and 5.1
• Microsoft FORTRAN for OS/2, versions 4.1, 5.0 and 5.1

SYMPTOMS

When using the Microsoft FORTRAN Compiler version 4.1, 5.0, or 5.1, if FORTRAN code attempts to access a C float function, incorrect results may be returned from the C function.

CAUSE

FORTRAN expects a DOUBLE PRECISION number to be returned from both REAL and DOUBLE PRECISION functions. C functions of type float return only 4 bytes, and the FORTRAN run-time incorrectly interprets the value stored in the floating accumulator (__fac) as an 8-byte number.

RESOLUTION

This problem can be avoided by declaring the C function as type double, or by declaring the C function with the _fortran attribute.

MORE INFORMATION

Sample Code #1

The following code reproduces the problem:

FORTRAN Code

       interface to real function cthing[c](dog)
       real dog[value]
       end

       real dog
       real cthing,r
       dog=2
       r=cthing(dog)
       write(*,*) r
       end

C Code

#include <stdio.h>

   float cat;
   printf("%f \n",dog);
   cat=3.0;
   printf("%f \n",cat);
   return cat;

Output: Incorrect Results Generated Here

2.000000 3.000000

   0.000000E+00

Sample Code #2

The following solution declares the C function with the _fortran attribute, and generates the expected output:

FORTRAN Code

       real dog,r
       real cthing
       dog=2.0
       r=cthing(dog)
       write(*,*) r
       end

C Code

#include <stdio.h>

  float cat;
  printf("%f \n",*dog);
  cat=3.0;
  printf("%f \n",cat);
  return cat;

Output

2.000000 3.000000

      3.000000